Tails of risk

By most accounts, the current financial crisis is epochal — the worst since the Depression. Still, it is a truism in asset management: look after the downside and the returns will look after themselves. But how well do asset managers look after the downside?

In the current market meltdown, evidence of failure isn’t hard to find. There are the bond-rating agencies using limited data to rate subprime mortgages bundled up as securities. There’s insurance giant AIG, which sold under-collateralized credit default swaps. There are the investment banks with leverage of over 30:1.

Could these things have been prevented? It depends on how you measure risk.

French business school EDHEC recently surveyed asset managers about portfolio construction. Generally, a portfolio is optimized to either weed out diversifiable risk along an efficient frontier, or maximize returns at a specified level of risk.

Risk is usually understood as standard deviation, or the variability of investment returns. A simple example for the first optimization model would be a 60/40 stock/bond portfolio that returns 9% with a risk of 8% — annual returns would vary 24% on the upside or the downside most of the time. That return profile spans three standard deviations, which cover about 95% of return probabilities. However, there is a problem with this risk measure, as EDHEC points out.

EDHEC looked at three approaches to risk: absolute, relative and extreme risk.

The goal is a positive return, year in and year out. Most fund managers in the survey have absolute risk objectives, defined by volatility (or standard deviation) as well as value-at-risk (VaR). VaR attempts to measure tail risk: the return probabilities that happen only 5% of the time.

Less popular is relative risk, which involves tracking error, the possibility of underperforming a benchmark, either a stock benchmark such as the S&P TSX Composite or one customized for various asset classes.

Finally, extreme risk involves protection against unusual events — the fat tails of a return distribution. Investment returns are not normally distributed, which means they don’t trace a classic bell curve pattern. There are outliers at each end of the curve and their frequency can drive VaR analysis, or potential maximum daily loss. Thus, the survey notes, deviations from normal distributions have to be incorporated into risk measures. But fewer than 20% of the fund managers surveyed do so.

Extreme risk measures are not limited to VaR. Managers may also do stress testing or Monte Carlo simulations, among other things.

Another part of risk management is the quality of inputs. Most managers use a sample drawn from a covariance matrix to estimate how different assets move, without adjusting with more sophisticated techniques.

On the other hand, with optimization models that attempt to estimate expected returns, sometimes the output is all over the map, so some constraint has to be introduced. But, as in other measures, not many fund managers have adopted higher-order techniques to guide in refining the technique.

The survey also canvassed the views of practitioners themselves. Revealingly, while many acknowledged that they were not using the latest statistical techniques, a few suggested they prefer to make qualitative modifications because they “do not trust to the optimizer or do not know quant tools.”

Indeed, half of respondents suggested that a lack of knowledge was the key reason for managers not using these techniques.

(03/25/09)